Introduction to Process Mining

Setup

You need to install several R packages from the BupaRverse.

install.packages("bupaR")
install.packages("edeaR")
install.packages("eventdataR")
install.packages("processmapR")
install.packages("processanimateR")
install.packages("petrinetR")

Slide 8

library(bupaR)

## Warning: package 'bupaR' was built under R version 4.1.2

## 
## Attaching package: 'bupaR'

## The following object is masked from 'package:stats':
## 
##     filter

## The following object is masked from 'package:utils':
## 
##     timestamp

bupaR autmaticaly loads eventdataR

#library(eventdataR)

Retrieve example datasets from the eventdataR

When loading the bupaR package, the traffic_fines dataset is also loaded into memory.

Print the case long format table (slide 8)

traffic_fines

## # A tibble: 34,724 × 18
##    case_id activity        lifecycle resource timestamp           amount article
##    <chr>   <fct>           <fct>     <fct>    <dttm>               <dbl>   <int>
##  1 A1      Create Fine     complete  561      2006-07-24 00:00:00    350     157
##  2 A1      Send Fine       complete  <NA>     2006-12-05 00:00:00     NA      NA
##  3 A100    Create Fine     complete  561      2006-08-02 00:00:00    350     157
##  4 A100    Send Fine       complete  <NA>     2006-12-12 00:00:00     NA      NA
##  5 A100    Insert Fine No… complete  <NA>     2007-01-15 00:00:00     NA      NA
##  6 A100    Add penalty     complete  <NA>     2007-03-16 00:00:00    715      NA
##  7 A100    Send for Credi… complete  <NA>     2009-03-30 00:00:00     NA      NA
##  8 A10000  Create Fine     complete  561      2007-03-09 00:00:00    360     157
##  9 A10000  Send Fine       complete  <NA>     2007-07-17 00:00:00     NA      NA
## 10 A10000  Insert Fine No… complete  <NA>     2007-08-02 00:00:00     NA      NA
## # … with 34,714 more rows, and 11 more variables: dismissal <chr>,
## #   expense <dbl>, lastsent <chr>, matricola <chr>, notificationtype <chr>,
## #   paymentamount <dbl>, points <int>, totalpaymentamount <chr>,
## #   vehicleclass <chr>, activity_instance_id <chr>, .order <int>

We have to turn the traffic_fines dataset into an eventlog/log class using the simple_eventlog function.

log_traffic_fines <- simple_eventlog(traffic_fines)

## Warning: `simple_eventlog()` was deprecated in bupaR 5.0.0.
## Please use `activitylog()` instead.

For this dataset, you can also just add the “log” class to the R object.

log_traffic_fines <- traffic_fines
class(log_traffic_fines) <- c("log",class(log_traffic_fines))

1. Hands-on training From the same eventdataR package, retrieve the datasets:

1. sepsis
2. hospital_billing
3. hospital
4. patients

Slide 9

Print the case long format table (slide 9)

case_list(log_traffic_fines)

## # A tibble: 10,000 × 3
##    case_id trace                                                        trace_id
##    <chr>   <chr>                                                           <dbl>
##  1 A2127   Create Fine,Payment                                                 1
##  2 A15     Create Fine,Send Fine,Insert Fine Notification,Add penalty,…       25
##  3 A1820   Create Fine,Payment                                                 1
##  4 A22     Create Fine,Send Fine,Insert Fine Notification,Add penalty,…       19
##  5 A1451   Create Fine,Send Fine,Insert Fine Notification,Add penalty,…       25
##  6 A23     Create Fine,Send Fine,Insert Fine Notification,Add penalty,…       20
##  7 A24     Create Fine,Send Fine                                               7
##  8 A25     Create Fine,Send Fine,Insert Fine Notification,Add penalty,…       25
##  9 A1452   Create Fine,Send Fine,Insert Fine Notification,Add penalty,…       25
## 10 A1453   Create Fine,Send Fine,Insert Fine Notification,Add penalty,…       25
## # … with 9,990 more rows

A more detailed view on lists:

cases(log_traffic_fines)

## # A tibble: 10,000 × 10
##    case_id trace_length number_of_activ… start_timestamp     complete_timestamp 
##    <chr>          <int>            <int> <dttm>              <dttm>             
##  1 A1                 2                2 2006-07-24 00:00:00 2006-12-05 00:00:00
##  2 A100               5                5 2006-08-02 00:00:00 2009-03-30 00:00:00
##  3 A10000             5                5 2007-03-09 00:00:00 2008-09-09 00:00:00
##  4 A10001             6                6 2007-03-19 00:00:00 2007-09-24 00:00:00
##  5 A10004             5                5 2007-03-20 00:00:00 2009-03-30 00:00:00
##  6 A10005             2                2 2007-03-20 00:00:00 2007-03-21 00:00:00
##  7 A10007             2                2 2007-03-20 00:00:00 2007-03-23 00:00:00
##  8 A10008             5                5 2007-03-20 00:00:00 2009-03-30 00:00:00
##  9 A10009             6                5 2007-03-20 00:00:00 2007-10-31 00:00:00
## 10 A1001              5                5 2006-08-02 00:00:00 2009-03-30 00:00:00
## # … with 9,990 more rows, and 5 more variables: trace <chr>, trace_id <dbl>,
## #   duration <drtn>, first_activity <fct>, last_activity <fct>

Rapid glance to the cases’ length distribution

case_length_dist_tf <- sapply(strsplit(case_list(log_traffic_fines)$trace, ","), length) 

table(case_length_dist_tf)

## case_length_dist_tf
##    2    3    4    5    6    7    8    9 
## 5318   42    5 4031  542   10    3   49

barplot(table(case_length_dist_tf))

quantile(case_length_dist_tf,probs = seq(0, 1, 0.1))

##   0%  10%  20%  30%  40%  50%  60%  70%  80%  90% 100% 
##    2    2    2    2    2    2    5    5    5    5    9

log_traffic_fines %>%
  durations(units="hours") %>% # duration of the cases
  as.data.frame(units="hours") %>% # cases and their duration
  ggplot(aes(x=duration))+
  geom_density(color="#1F77b4",
                 fill="#17BECF",
                 alpha=.5)+
  theme_minimal()

## Don't know how to automatically pick scale for object of type difftime. Defaulting to continuous.

Print the trace table (slide 9)

trace_list(log_traffic_fines)

## # A tibble: 44 × 3
##    trace                                       absolute_freque… relative_freque…
##    <chr>                                                  <int>            <dbl>
##  1 Create Fine,Payment                                     3428           0.343 
##  2 Create Fine,Send Fine,Insert Fine Notifica…             3273           0.327 
##  3 Create Fine,Send Fine                                   1890           0.189 
##  4 Create Fine,Send Fine,Insert Fine Notifica…              758           0.0758
##  5 Create Fine,Send Fine,Insert Fine Notifica…              250           0.025 
##  6 Create Fine,Send Fine,Insert Fine Notifica…              151           0.0151
##  7 Create Fine,Send Fine,Insert Fine Notifica…               78           0.0078
##  8 Create Fine,Send Fine,Payment                             37           0.0037
##  9 Create Fine,Send Fine,Insert Fine Notifica…               24           0.0024
## 10 Create Fine,Send Fine,Insert Fine Notifica…               13           0.0013
## # … with 34 more rows

2. Hands-on training Display the case and trace tables as well as the cases’ lengths distribution for the following datasets:

1. sepsis
2. hospital_billing
3. hospital
4. patients

For instance, here are the results for the sepsis log.

## # A tibble: 15,214 × 34
##    case_id activity  lifecycle resource timestamp             age   crp diagnose
##    <chr>   <fct>     <fct>     <fct>    <dttm>              <int> <dbl> <chr>   
##  1 A       ER Regis… complete  A        2014-10-22 11:15:41    85    NA A       
##  2 A       Leucocyt… complete  B        2014-10-22 11:27:00    NA    NA <NA>    
##  3 A       CRP       complete  B        2014-10-22 11:27:00    NA   210 <NA>    
##  4 A       LacticAc… complete  B        2014-10-22 11:27:00    NA    NA <NA>    
##  5 A       ER Triage complete  C        2014-10-22 11:33:37    NA    NA <NA>    
##  6 A       ER Sepsi… complete  A        2014-10-22 11:34:00    NA    NA <NA>    
##  7 A       IV Liquid complete  A        2014-10-22 14:03:47    NA    NA <NA>    
##  8 A       IV Antib… complete  A        2014-10-22 14:03:47    NA    NA <NA>    
##  9 A       Admissio… complete  D        2014-10-22 14:13:19    NA    NA <NA>    
## 10 A       CRP       complete  B        2014-10-24 09:00:00    NA  1090 <NA>    
## # … with 15,204 more rows, and 26 more variables: diagnosticartastrup <chr>,
## #   diagnosticblood <chr>, diagnosticecg <chr>, diagnosticic <chr>,
## #   diagnosticlacticacid <chr>, diagnosticliquor <chr>, diagnosticother <chr>,
## #   diagnosticsputum <chr>, diagnosticurinaryculture <chr>,
## #   diagnosticurinarysediment <chr>, diagnosticxthorax <chr>, disfuncorg <chr>,
## #   hypotensie <chr>, hypoxie <chr>, infectionsuspected <chr>, infusion <chr>,
## #   lacticacid <chr>, leucocytes <chr>, oligurie <chr>, …

## # A tibble: 1,050 × 3
##    case_id trace                                                        trace_id
##    <chr>   <chr>                                                           <dbl>
##  1 XJ      ER Registration,ER Triage,ER Sepsis Triage,LacticAcid,Leuco…      496
##  2 I       ER Registration,ER Triage,ER Sepsis Triage,IV Antibiotics,L…      217
##  3 WEA     ER Registration,ER Triage,ER Sepsis Triage,CRP,Leucocytes,I…      147
##  4 OT      ER Registration,ER Triage,ER Sepsis Triage,LacticAcid,Leuco…      470
##  5 ULA     ER Registration,IV Liquid,ER Triage,Leucocytes,CRP,LacticAc…      772
##  6 BV      ER Registration,ER Triage,ER Sepsis Triage,CRP,LacticAcid,L…      110
##  7 MY      ER Registration,ER Triage,ER Sepsis Triage,CRP,LacticAcid,L…       74
##  8 ZS      ER Registration,ER Triage,ER Sepsis Triage,Leucocytes,CRP,L…      625
##  9 SGA     ER Registration,ER Triage,CRP,LacticAcid,Leucocytes,ER Seps…       37
## 10 QKA     ER Registration,ER Triage,ER Sepsis Triage,Leucocytes,CRP,L…      623
## # … with 1,040 more rows

## # A tibble: 1,050 × 10
##    case_id trace_length number_of_activ… start_timestamp     complete_timestamp 
##    <chr>          <int>            <int> <dttm>              <dttm>             
##  1 <NA>              24               10 2014-11-10 01:45:19 2014-11-22 14:30:00
##  2 A                 22               10 2014-10-22 11:15:41 2014-11-02 15:15:00
##  3 AA                 8                8 2014-12-03 09:06:44 2014-12-03 14:28:01
##  4 AAA               11               11 2014-11-19 03:16:21 2014-11-28 16:15:17
##  5 AB                 8                8 2014-02-16 09:55:43 2014-02-16 13:52:06
##  6 ABA               17               10 2014-10-12 11:22:24 2014-10-18 16:15:00
##  7 AC                13               11 2014-09-24 15:39:13 2014-12-02 18:47:17
##  8 ACA                8                8 2014-09-22 10:00:00 2014-09-22 10:45:56
##  9 AD                29               12 2013-12-19 09:04:38 2014-05-04 18:34:00
## 10 ADA               24               10 2014-06-21 07:14:16 2014-07-06 20:58:00
## # … with 1,040 more rows, and 5 more variables: trace <chr>, trace_id <dbl>,
## #   duration <drtn>, first_activity <fct>, last_activity <fct>

## case_length_dist_se
##   3   4   5   6   7   8   9  10  11  12  13  14  15  16  17  18  19  20  21  22 
##  35   1  61  32  27 100  41  50  50  81  95  79  74  69  42  31  21  19  21  15 
##  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  43 
##  15   8   6   3   6   4   9   8   5   4   5   2   2   3   3   1   2   2   2   1 
##  45  48  51  52  59  61  66  84  88 118 170 185 
##   2   1   1   2   1   2   1   1   1   1   1   1

##   0%  10%  20%  30%  40%  50%  60%  70%  80%  90% 100% 
##    3    6    8   10   12   13   14   16   18   23  185

## Don't know how to automatically pick scale for object of type difftime. Defaulting to continuous.

Slide 10

The print function displays a summary of the log. The same result is provided by just typinf the log’s object name in the Rconsole.

print(log_traffic_fines)

## # Log of 34724 events consisting of:
## 44 traces 
## 10000 cases 
## 34724 instances of 11 activities 
## 1 resource 
## Events occurred from 2006-06-17 until 2012-03-26 
##  
## # Variables were mapped as follows:
## Case identifier:     case_id 
## Activity identifier:     activity 
## Resource identifier:     resource_id 
## Activity instance identifier:    activity_instance_id 
## Timestamp:           timestamp 
## Lifecycle transition:        lifecycle_id 
## 
## # A tibble: 34,724 × 20
##    case_id activity        lifecycle resource timestamp           amount article
##    <chr>   <fct>           <fct>     <fct>    <dttm>               <dbl>   <int>
##  1 A1      Create Fine     complete  561      2006-07-24 00:00:00    350     157
##  2 A1      Send Fine       complete  <NA>     2006-12-05 00:00:00     NA      NA
##  3 A100    Create Fine     complete  561      2006-08-02 00:00:00    350     157
##  4 A100    Send Fine       complete  <NA>     2006-12-12 00:00:00     NA      NA
##  5 A100    Insert Fine No… complete  <NA>     2007-01-15 00:00:00     NA      NA
##  6 A100    Add penalty     complete  <NA>     2007-03-16 00:00:00    715      NA
##  7 A100    Send for Credi… complete  <NA>     2009-03-30 00:00:00     NA      NA
##  8 A10000  Create Fine     complete  561      2007-03-09 00:00:00    360     157
##  9 A10000  Send Fine       complete  <NA>     2007-07-17 00:00:00     NA      NA
## 10 A10000  Insert Fine No… complete  <NA>     2007-08-02 00:00:00     NA      NA
## # … with 34,714 more rows, and 13 more variables: dismissal <chr>,
## #   expense <dbl>, lastsent <chr>, matricola <chr>, notificationtype <chr>,
## #   paymentamount <dbl>, points <int>, totalpaymentamount <chr>,
## #   vehicleclass <chr>, activity_instance_id <int>, .order <int>,
## #   resource_id <chr>, lifecycle_id <chr>

To get the number of events:

n_events(log_traffic_fines)

## [1] 34724

To get the number of cases:

n_cases(log_traffic_fines)

## [1] 10000

To get the number of traces:

n_traces(log_traffic_fines)

## [1] 44

To get the number of activities:

n_activities(log_traffic_fines)

## [1] 11

Recall the formula for Average case length:

\[\displaystyle\frac{Event}{Cases}=\frac{\mathrm{n\_events(log\_traffic\_fines)}}{\mathrm{n\_cases(log\_traffic\_fines)}}\]

n_events(log_traffic_fines)/n_cases(log_traffic_fines)

## [1] 3.4724

\(\displaystyle\frac{Event}{Cases}=3.4724\)

Recall the formula for Process uniformity:

\[\displaystyle 1−\frac{Traces-1}{Cases}=1-\frac{\mathrm{n\_traces(log\_traffic\_fines)-1}}{\mathrm{n\_cases(log\_traffic\_fines)}}\]

1-(n_traces(log_traffic_fines)-1)/n_cases(log_traffic_fines)

## [1] 0.9957

\(\displaystyle\frac{Event}{Cases}=0.9957\)

3. Hands-on training Compute the average case length and the process uniformity for the following datasets:

1. sepsis
2. hospital_billing
3. hospital
4. patients

Slide 13

Compute and display the Precedence Matrix for absolute frequencies.

process_matrix(log_traffic_fines)

## # A tibble: 47 × 3
##    antecedent      consequent                                n
##    <fct>           <fct>                                 <dbl>
##  1 Add penalty     Insert Date Appeal to Prefecture         41
##  2 Add penalty     Notify Result Appeal to Offender          3
##  3 Add penalty     Payment                                1117
##  4 Add penalty     Receive Result Appeal from Prefecture    15
##  5 Add penalty     Send Appeal to Prefecture               171
##  6 Add penalty     Send for Credit Collection             3288
##  7 Appeal to Judge Add penalty                              13
##  8 Appeal to Judge End                                       5
##  9 Appeal to Judge Insert Date Appeal to Prefecture          1
## 10 Create Fine     Payment                                3443
## # … with 37 more rows

plot(process_matrix(log_traffic_fines))

It is a ggplot object that can be customized.

plot(process_matrix(log_traffic_fines))+
  geom_tile(aes(fill=n))+
  geom_text(aes(label=n),color="white",size=2.5)+
  ggtitle("Traffic fines process map")+
  scale_fill_viridis_c(option="turbo")+
  theme_light()+
  theme(axis.text.x=element_text(angle=45,
                                 hjust=1,
                                 size=8),
        axis.text.y=element_text(size=8))

## Scale for 'fill' is already present. Adding another scale for 'fill', which
## will replace the existing scale.

In addition to the absolute frequencies, other quantities can be computed:

• “absolute”,
• “relative”,
• “absolute-case”,
• “relative-case”,
• “relative-antecedent”,
• “relative-consequent”,

where absolute, relative (percentage of activity instances) or relative_case (percentage of cases the activity occurs in).

plot(process_matrix(log_traffic_fines,type=frequency(value="relative")))+geom_tile(aes(fill = rel_n))+geom_text(aes(label=round(rel_n * 100, 2)),color="white",size=2.5)

plot(process_matrix(log_traffic_fines,type=frequency(value="absolute-case")))+geom_tile(aes(fill = n_cases))+geom_text(aes(label=n_cases),color="white",size=2.5)

4. Hands-on training Compute and display precedence matrices for the following datasets:

1. sepsis
2. hospital_billing
3. hospital
4. patients

Slide 17

The percentage coverage of the trace to explore. Default is 20% most (in)frequent

trace_explorer(log_traffic_fines, coverage=1)

Instead of setting coverage, you can set an exact number of traces. Should be an integer larger than 0.

trace_explorer(log_traffic_fines, n_traces = 10)

library(TraMineR)

## Warning: package 'TraMineR' was built under R version 4.1.2

## 
## TraMineR stable version 2.2-4 (Built: 2022-06-10)

## Website: http://traminer.unige.ch

## Please type 'citation("TraMineR")' for citation information.

Plot selected cases.

seqplot(s_tf,
        type="i",  
        cex.axis=.8,
        cex.legend=.8,
        legend.prop = .44, 
        with.legend = "right",
)

Other types :

• d, for state distribution plots
• dH, for chronograms with overlayed entropy line
• f, for sequence frequency plots
• i, for selected sequence index plots
• I, for whole set index plots
• Ht, for transversal entropy plots
• ms, for plotting the sequence of modal states
• mt, for mean times plots
• pc, for parallel coordinate plots

seqplot(s_tf,
        type="dH",  
        cex.axis=.8,
        cex.legend=.8,legend.prop = .44, with.legend = "right",
)

seqplot(s_tf,
        type="Ht",  
        cex.axis=.8,
        cex.legend=.8,
        legend.prop = .44, 
        with.legend = "right",
)

seqplot(s_tf,
        type="pc",  
        cex.axis=.8,
        cex.legend=.8,
        legend.prop = .44, 
        with.legend = "right",
)

To get the state frequencies use, the seqstatd function.

seqstatd(s_tf, weighted=FALSE)

##                                                                           [State frequencies]
##                                       [1]  [2]     [3]     [4]     [5]    [6]
## Add penalty                             0 0.00 0.00000 0.95000 0.04466 0.0248
## Appeal to Judge                         0 0.00 0.00000 0.00302 0.00000 0.0000
## Create Fine                             1 0.00 0.00000 0.00000 0.00000 0.0000
## Insert Date Appeal to Prefecture        0 0.00 0.00043 0.04052 0.00906 0.0000
## Insert Fine Notification                0 0.00 0.98804 0.00194 0.00000 0.0000
## Notify Result Appeal to Offender        0 0.00 0.00000 0.00000 0.00022 0.0033
## Payment                                 0 0.34 0.00876 0.00431 0.23581 0.4636
## Receive Result Appeal from Prefecture   0 0.00 0.00000 0.00000 0.00043 0.0199
## Send Appeal to Prefecture               0 0.00 0.00000 0.00022 0.00367 0.3361
## Send Fine                               0 0.66 0.00278 0.00000 0.00000 0.0000
## Send for Credit Collection              0 0.00 0.00000 0.00000 0.70615 0.1523
##                                         [7]   [8]   [9]
## Add penalty                           0.032 0.058 0.000
## Appeal to Judge                       0.000 0.019 0.082
## Create Fine                           0.000 0.000 0.000
## Insert Date Appeal to Prefecture      0.000 0.000 0.000
## Insert Fine Notification              0.000 0.000 0.000
## Notify Result Appeal to Offender      0.145 0.808 0.000
## Payment                               0.097 0.058 0.490
## Receive Result Appeal from Prefecture 0.661 0.000 0.000
## Send Appeal to Prefecture             0.048 0.000 0.061
## Send Fine                             0.000 0.000 0.000
## Send for Credit Collection            0.016 0.058 0.367
## 
##                                                                            [Valid states]
##                                         [1]   [2]  [3]  [4]  [5] [6] [7] [8]
## N                                     10000 10000 4682 4640 4635 604  62  52
##                                       [9]
## N                                      49
## 
##                                                                            [Entropy index]
##                                       [1]  [2]  [3]   [4]  [5] [6]  [7]  [8]
## H                                       0 0.27 0.03 0.097 0.33 0.5 0.46 0.31
##                                        [9]
## H                                     0.46

5. Hands-on training Use the trace_explorer and compute entropies for the following datasets:

1. sepsis
2. hospital_billing
3. hospital
4. patients

Slide 21

library(heuristicsmineR)

## Warning: package 'heuristicsmineR' was built under R version 4.1.2

## 
## Attaching package: 'heuristicsmineR'

## The following object is masked from 'package:processmapR':
## 
##     precedence_matrix

process_map(L_heur_1)

You can easily filter the data to map using the %>% operator.

pm<-L_heur_1 %>%
  filter_activity_frequency(percentage=1) %>%
  filter_trace_frequency(percentage=1) %>%
  process_map(sec_nodes=frequency(value="absolute"),
              type=performance(FUN=mean,
                               units="hours",
                               color_scale="Greens",
                               color_edges="Greens"),
              unit="hours")
pm

pm<-L_heur_1 %>%
  filter_activity_frequency(percentage=.5) %>%
  filter_trace_frequency(percentage=.5) %>%
  process_map(sec_nodes=frequency(value="absolute"),
              type=performance(FUN=mean,
                               units="hours",
                               color_scale="Greens",
                               color_edges="Greens"),
              unit="hours")
pm

6. Hands-on training Create those maps for the following datasets:

1. sepsis
2. hospital_billing
3. hospital
4. patients

Slide 25

Full dataset

process_map(log_traffic_fines, rankdir = "TB")

Filtering to get a 90% coverage of activity instances.

log_traffic_fines %>%
  filter_trace_frequency(percentage=.9) %>%
process_map(rankdir = "TB")

7. Hands-on training Create those maps for the following datasets:

1. sepsis
2. hospital_billing
3. hospital
4. patients

Slide 36

library("viridisLite")
m<-log_sepsis %>%
  dependency_matrix(threshold=0) %>%
  (reshape2::melt)
m[m$value!=0,] %>%
  ggplot(aes(consequent, antecedent, label=round(value,2)))+
  geom_tile(aes(fill=value), color="grey27")+
  geom_text(color="white",size=2.5)+
  theme_light()+
  theme(axis.text.x=element_text(angle=45,
                                 hjust=1,
                                 size=8),
        axis.text.y=element_text(size=8))

library("viridisLite")
m<-log_sepsis %>%
  dependency_matrix(threshold=.9) %>%
  (reshape2::melt)
m[m$value!=0,] %>%
  ggplot(aes(consequent, antecedent, label=round(value,2)))+
  geom_tile(aes(fill=value), color="grey27")+
  geom_text(color="white",size=2.5)+
  theme_light()+
  theme(axis.text.x=element_text(angle=45,
                                 hjust=1,
                                 size=8),
        axis.text.y=element_text(size=8))

Slide 39

To compute the process map, first:

sepsis

## # A tibble: 15,214 × 34
##    case_id activity  lifecycle resource timestamp             age   crp diagnose
##    <chr>   <fct>     <fct>     <fct>    <dttm>              <int> <dbl> <chr>   
##  1 A       ER Regis… complete  A        2014-10-22 11:15:41    85    NA A       
##  2 A       Leucocyt… complete  B        2014-10-22 11:27:00    NA    NA <NA>    
##  3 A       CRP       complete  B        2014-10-22 11:27:00    NA   210 <NA>    
##  4 A       LacticAc… complete  B        2014-10-22 11:27:00    NA    NA <NA>    
##  5 A       ER Triage complete  C        2014-10-22 11:33:37    NA    NA <NA>    
##  6 A       ER Sepsi… complete  A        2014-10-22 11:34:00    NA    NA <NA>    
##  7 A       IV Liquid complete  A        2014-10-22 14:03:47    NA    NA <NA>    
##  8 A       IV Antib… complete  A        2014-10-22 14:03:47    NA    NA <NA>    
##  9 A       Admissio… complete  D        2014-10-22 14:13:19    NA    NA <NA>    
## 10 A       CRP       complete  B        2014-10-24 09:00:00    NA  1090 <NA>    
## # … with 15,204 more rows, and 26 more variables: diagnosticartastrup <chr>,
## #   diagnosticblood <chr>, diagnosticecg <chr>, diagnosticic <chr>,
## #   diagnosticlacticacid <chr>, diagnosticliquor <chr>, diagnosticother <chr>,
## #   diagnosticsputum <chr>, diagnosticurinaryculture <chr>,
## #   diagnosticurinarysediment <chr>, diagnosticxthorax <chr>, disfuncorg <chr>,
## #   hypotensie <chr>, hypoxie <chr>, infectionsuspected <chr>, infusion <chr>,
## #   lacticacid <chr>, leucocytes <chr>, oligurie <chr>, …

log_sepsis <- simple_eventlog(sepsis)

process_map(log_sepsis, rankdir = "TB")

Quite difficult to use, we need to mine it.

Slide 40

To compute the causal net, first:

Dependency matrix

sepsis

## # A tibble: 15,214 × 34
##    case_id activity  lifecycle resource timestamp             age   crp diagnose
##    <chr>   <fct>     <fct>     <fct>    <dttm>              <int> <dbl> <chr>   
##  1 A       ER Regis… complete  A        2014-10-22 11:15:41    85    NA A       
##  2 A       Leucocyt… complete  B        2014-10-22 11:27:00    NA    NA <NA>    
##  3 A       CRP       complete  B        2014-10-22 11:27:00    NA   210 <NA>    
##  4 A       LacticAc… complete  B        2014-10-22 11:27:00    NA    NA <NA>    
##  5 A       ER Triage complete  C        2014-10-22 11:33:37    NA    NA <NA>    
##  6 A       ER Sepsi… complete  A        2014-10-22 11:34:00    NA    NA <NA>    
##  7 A       IV Liquid complete  A        2014-10-22 14:03:47    NA    NA <NA>    
##  8 A       IV Antib… complete  A        2014-10-22 14:03:47    NA    NA <NA>    
##  9 A       Admissio… complete  D        2014-10-22 14:13:19    NA    NA <NA>    
## 10 A       CRP       complete  B        2014-10-24 09:00:00    NA  1090 <NA>    
## # … with 15,204 more rows, and 26 more variables: diagnosticartastrup <chr>,
## #   diagnosticblood <chr>, diagnosticecg <chr>, diagnosticic <chr>,
## #   diagnosticlacticacid <chr>, diagnosticliquor <chr>, diagnosticother <chr>,
## #   diagnosticsputum <chr>, diagnosticurinaryculture <chr>,
## #   diagnosticurinarysediment <chr>, diagnosticxthorax <chr>, disfuncorg <chr>,
## #   hypotensie <chr>, hypoxie <chr>, infectionsuspected <chr>, infusion <chr>,
## #   lacticacid <chr>, leucocytes <chr>, oligurie <chr>, …

log_sepsis <- simple_eventlog(sepsis)

library(heuristicsmineR)
plot(dependency_matrix(log_sepsis))+geom_raster(aes(fill = dep))+geom_text(aes(label=round(dep, 2)),color="white",size=2.5)

The option dependency_type enables to tune the computation of the dependency matrix. For instance, dependency_type=dependency_type_fhm(...) where ... stands for:

dependency_type_fhm(
  threshold_dependency = 0.9,
  threshold_l1 = threshold_dependency,
  threshold_l2 = threshold_dependency,
  threshold_frequency = 0,
  all_connected = FALSE,
  endpoints_connected = FALSE
)

with

• threshold_dependency A dependency threshold, usually in the interval [0,1], filtering out dependencies below the threshold.

• threshold_l1 A dependency threshold, usually in the interval [0,1], filtering out self-loop dependencies below the threshold.

• threshold_l2 A dependency threshold, usually in the interval [0,1], filtering out length-two loop dependencies below the threshold.

• threshold_frequency An absolute frequency threshold filtering dependencies which are observed infrequently.

• all_connected If TRUE the best antecedent and consequent (as determined by the dependency measure) are going to be added regardless of the threshold value.

• endpoints_connected If TRUE the start/end activity is added as antecedent/consequent when an activity would not be connected according to the threshold value.

library("viridisLite")
m<-log_sepsis %>%
  dependency_matrix(threshold=.8) %>%
  (reshape2::melt)
m[m$value!=0,] %>%
  ggplot(aes(consequent, antecedent, label=round(value,2)))+
  geom_tile(aes(fill=value), color="grey27")+
  scale_fill_gradientn(colors=viridis(256),
                       name="Dependency\nmeasure")+
  theme_light()+
  theme(axis.text.x=element_text(angle=45,
                                 hjust=1,
                                 size=8),
        axis.text.y=element_text(size=8))

cn_sep0<-causal_net(log_sepsis)

## Warning in check_dependencies(dependencies): Activities [Release E] have neither
## an antecedent or consequent in the supplied dependency matrix.Consider using
## the `all_connected` or `endpoints_connected` parameter when generating the
## dependency matrix.

cn_sep0

## Nodes
## # A tibble: 18 × 12
##    act       from_id     n n_distinct_cases bindings_input bindings_output label
##    <chr>       <int> <dbl>            <dbl> <list>         <list>          <chr>
##  1 Admissio…       1   117              110 <list [1]>     <list [0]>      "Adm…
##  2 Admissio…       2  1182              800 <list [3]>     <list [4]>      "Adm…
##  3 CRP             3  3262             1007 <list [3]>     <list [7]>      "CRP…
##  4 End             4  1050             1050 <list [51]>    <list [0]>      "End"
##  5 ER Regis…       5  1050             1050 <list [1]>     <list [1]>      "ER …
##  6 ER Sepsi…       6  1049             1049 <list [1]>     <list [3]>      "ER …
##  7 ER Triage       7  1053             1050 <list [1]>     <list [1]>      "ER …
##  8 IV Antib…       8   823              823 <list [1]>     <list [4]>      "IV …
##  9 IV Liquid       9   753              753 <list [3]>     <list [1]>      "IV …
## 10 LacticAc…      10  1466              860 <list [1]>     <list [2]>      "Lac…
## 11 Leucocyt…      11  3383             1012 <list [3]>     <list [6]>      "Leu…
## 12 Release A      12   671              671 <list [5]>     <list [2]>      "Rel…
## 13 Release B      13    56               56 <list [1]>     <list [1]>      "Rel…
## 14 Release C      14    25               25 <list [1]>     <list [1]>      "Rel…
## 15 Release D      15    24               24 <list [1]>     <list [2]>      "Rel…
## 16 Release E      16     6                6 <list [0]>     <list [0]>      "Rel…
## 17 Return ER      17   294              294 <list [2]>     <list [1]>      "Ret…
## 18 Start          18  1050             1050 <list [0]>     <list [5]>      "Sta…
## # … with 5 more variables: color_level <dbl>, shape <chr>, fontcolor <chr>,
## #   color <chr>, tooltip <chr>
## Edges
## # A tibble: 36 × 8
##    antecedent       consequent     dep from_id to_id     n label penwidth
##    <chr>            <chr>        <dbl>   <int> <int> <dbl> <chr>    <dbl>
##  1 IV Antibiotics   Admission IC 0.979       8     1    98 98        1.17
##  2 Admission NC     Admission NC 0.994       2     2   381 381       1.64
##  3 IV Antibiotics   Admission NC 0.990       8     2   690 690       2.17
##  4 CRP              CRP          0.997       3     3  2251 2251      4.81
##  5 Start            CRP          0.909      18     3  1006 1006      2.70
##  6 Admission NC     End          0.933       2     4   129 129       1.22
##  7 CRP              End          0.976       3     4   241 241       1.41
##  8 ER Sepsis Triage End          0.98        6     4   226 226       1.38
##  9 IV Antibiotics   End          0.989       8     4   124 124       1.21
## 10 IV Liquid        End          0.923       9     4   752 752       2.27
## # … with 26 more rows

Consider using the all_connected or endpoints_connected parameter when generating the dependency matrix.

cn_sep<-causal_net(log_sepsis, all_connected = TRUE)
cn_sep

## Nodes
## # A tibble: 18 × 12
##    act       from_id     n n_distinct_cases bindings_input bindings_output label
##    <chr>       <int> <dbl>            <dbl> <list>         <list>          <chr>
##  1 Admissio…       1   117              110 <list [1]>     <list [1]>      "Adm…
##  2 Admissio…       2  1182              800 <list [3]>     <list [4]>      "Adm…
##  3 CRP             3  3262             1007 <list [3]>     <list [8]>      "CRP…
##  4 End             4  1050             1050 <list [54]>    <list [0]>      "End"
##  5 ER Regis…       5  1050             1050 <list [1]>     <list [1]>      "ER …
##  6 ER Sepsi…       6  1049             1049 <list [1]>     <list [3]>      "ER …
##  7 ER Triage       7  1053             1050 <list [1]>     <list [1]>      "ER …
##  8 IV Antib…       8   823              823 <list [1]>     <list [4]>      "IV …
##  9 IV Liquid       9   753              753 <list [3]>     <list [1]>      "IV …
## 10 LacticAc…      10  1466              860 <list [3]>     <list [2]>      "Lac…
## 11 Leucocyt…      11  3383             1012 <list [3]>     <list [6]>      "Leu…
## 12 Release A      12   671              671 <list [5]>     <list [2]>      "Rel…
## 13 Release B      13    56               56 <list [1]>     <list [1]>      "Rel…
## 14 Release C      14    25               25 <list [1]>     <list [1]>      "Rel…
## 15 Release D      15    24               24 <list [1]>     <list [2]>      "Rel…
## 16 Release E      16     6                6 <list [1]>     <list [1]>      "Rel…
## 17 Return ER      17   294              294 <list [2]>     <list [1]>      "Ret…
## 18 Start          18  1050             1050 <list [0]>     <list [5]>      "Sta…
## # … with 5 more variables: color_level <dbl>, shape <chr>, fontcolor <chr>,
## #   color <chr>, tooltip <chr>
## Edges
## # A tibble: 39 × 8
##    antecedent       consequent     dep from_id to_id     n label penwidth
##    <chr>            <chr>        <dbl>   <int> <int> <dbl> <chr>    <dbl>
##  1 IV Antibiotics   Admission IC 0.979       8     1    98 98        1.17
##  2 Admission NC     Admission NC 0.994       2     2   381 381       1.64
##  3 IV Antibiotics   Admission NC 0.990       8     2   690 690       2.17
##  4 CRP              CRP          0.997       3     3  2251 2251      4.81
##  5 Start            CRP          0.909      18     3  1006 1006      2.70
##  6 Admission NC     End          0.933       2     4   129 129       1.22
##  7 CRP              End          0.976       3     4   235 235       1.40
##  8 ER Sepsis Triage End          0.98        6     4   226 226       1.38
##  9 IV Antibiotics   End          0.989       8     4   124 124       1.21
## 10 IV Liquid        End          0.923       9     4   752 752       2.27
## # … with 29 more rows

8. Hands-on training Compute precedence matrices and causal nets for the following datasets:

1. traffic_fines
2. hospital_billing
3. hospital
4. patients

Slide 40

pn1<-(as.petrinet(cn_sep))
unique(pn1$flows$to[!pn1$flows$to %in% pn1$flows$from])

## [1] "p_in_4"

render_PN(pn1)

Place/Transition Connecting Degree (PTCD)

(n_flows(pn1)/n_places(pn1)+n_flows(pn1)/n_transitions(pn1))/2

## [1] 4.664809

9. Hands-on training Compute and display precedence matrices for the following datasets:

1. traffic_fines
2. hospital_billing
3. hospital
4. patients

Animate process maps

library(processanimateR)

## Warning: package 'processanimateR' was built under R version 4.1.2

animate_process(log_traffic_fines)

A more intricate example, from

library(processanimateR)
library(dplyr)
library(bupaR)
# Extract only the lacticacid measurements
lactic <- log_sepsis %>%
    mutate(lacticacid = as.numeric(lacticacid)) %>%
    filter_activity(c("LacticAcid")) %>%
    as.data.frame() %>%
    select("case" = case_id, 
            "time" =  timestamp, 
            value = lacticacid) # format needs to be 'case,time,value'

# Remove the measurement events from the sepsis log
sepsisBase <- log_sepsis %>%
    filter_activity(c("LacticAcid", "CRP", "Leucocytes", "Return ER",
                      "IV Liquid", "IV Antibiotics"), reverse = T) %>%
    filter_trace_frequency(percentage = 0.95)

# Animate with the secondary data frame `lactic`
animate_process(sepsisBase, 
                mode = "relative", 
                duration = 300,
                legend = "color", 
                mapping = token_aes(color = token_scale(lactic, 
                                                        scale = "linear", 
                                                        range = c("#fff5eb","#7f2704")))) 

10. Hands-on training Display animate the processes of the following logs:

1. traffic_fines
2. hospital_billing
3. hospital
4. patients